repeated error

[chylerbot-beep]

repeatedly shows this error

Ran into this error: Request failed: Stream decode error: Failed to parse streaming chunk: invalid type: null, expected a string at line 1 column 263: "{"id":"gen-1769651335-VwOF0cWjCzEzs1z6BkZF","provider":"AtlasCloud","model":"deepseek/deepseek-v3.2","object":"chat.completion.chunk","created":1769651335,"choices":[{"index":0,"delta":{"role":"assistant","content":null,"tool_calls":[{"function":{"arguments":null},"index":0,"id":"","type":"function"}]},"finish_reason":null,"native_finish_reason":null,"logprobs":null}],"system_fingerprint":null}".

Please retry if you think this is a transient or recoverable error.

[Earnest-Williams]

Same here and I'm using Codex for the backend.

[DOsinga]

this is fixed in #7076 - some models don't like no content when there is a toolcall

[PadraigOreilly251]

Hi, I´m running onto it in version 1.26 :

Ran into this error: Request failed: Stream decode error: Failed to parse streaming chunk: missing field choices at line 1 column 99: "{"error":{"code":500,"message":"Invalid diff: now finding less tool calls!","type":"server_error"}}".

Please retry if you think this is a transient or recoverable error.

[poof86]

Same here, trying it for the first time using LM Studio as a provider with model "Devstral small 2"

[PadraigOreilly251]

I've had problems with devstral small 2 in general(at least with unsloth's ggufs) both in Goose and Roo. I think Devstral is VERY particular about how it does tool calls and agentic in general. Something about turns, i think? Maybe it works better with mistral's own framework 🤷🏻‍♂️

On that line I'm still trying to set up a MLX server that does proper MCP calls. I've been stuck using llama.cpp and I think I'm potentially losing ~30% speed in the Mac because of that.

[DOsinga]

does anybody here have a diagnostics report? @PadraigOreilly251 - check the latest version where we have local models natively integrated

[PadraigOreilly251]

I´m testing it now, and I think it´s working a bit better.
I don´t know if it´s the "local" thing because I´m more like "techincally" local (I have several models loaded in different machines, and I access them via LAN and Tailscale). It might also be that I´m systematically disabling the "summon" extension, and I´m also adding to Goose Hints (with the intention to escalate to the system prompt if necessary) that it shouldn´t create subagents.

I will test it for longer and report

[DOsinga]

I'm going to close this - without a diagnostics report we can't really make progress on this. feel free to ping if we have something

[CezaryKlus]

This should be reopened. I am experiencing it with version 1.30 with Azure OpenAI provider

[PadraigOreilly251]

I have a couple of diagnostics reports if it helps